Computer Vision Method for Automatic Detection of Microstructure Defects of Concrete.

The search for structural and microstructural defects using simple human vision is associated with significant errors in determining voids, large pores, and violations of the integrity and compactness of particle packing in the micro- and macrostructure of concrete. Computer vision methods, in particular convolutional neural networks, have proven to be reliable tools for the automatic detection of defects during visual inspection of building structures. The study's objective is to create and compare computer vision algorithms that use convolutional neural networks to identify and analyze damaged sections in concrete samples from different structures. Networks of the following architectures were selected for operation: U-Net, LinkNet, and PSPNet. The analyzed images are photos of concrete samples obtained by laboratory tests to assess the quality in terms of the defection of the integrity and compactness of the structure. During the implementation process, changes in quality metrics such as macro-averaged precision, recall, and F1-score, as well as IoU (Jaccard coefficient) and accuracy, were monitored. The best metrics were demonstrated by the U-Net model, supplemented by the cellular automaton algorithm: precision = 0.91, recall = 0.90, F1 = 0.91, IoU = 0.84, and accuracy = 0.90. The developed segmentation algorithms are universal and show a high quality in highlighting areas of interest under any shooting conditions and different volumes of defective zones, regardless of their localization. The automatization of the process of calculating the damage area and a recommendation in the "critical/uncritical" format can be used to assess the condition of concrete of various types of structures, adjust the formulation, and change the technological parameters of production.

Detection and Dispersion Analysis of Water Globules in Oil Samples Using Artificial Intelligence Algorithms.

Fluid particle detection technology is of great importance in the oil and gas industry for improving oil-refining techniques and in evaluating the quality of refining equipment. The article discusses the process of creating a computer vision algorithm that allows the user to detect water globules in oil samples and analyze their sizes. The process of developing an algorithm based on the convolutional neural network (CNN) YOLOv4 is presented. For this study, our own empirical base was proposed, which comprised microphotographs of samples of raw materials and water-oil emulsions taken at various points and in different operating modes of an oil refinery. The number of images for training the neural network algorithm was increased by applying the authors' augmentation algorithm. The developed program makes it possible to detect particles in a fluid medium with the level of accuracy required by a researcher, which can be controlled at the stage of training the CNN. Based on the results of processing the output data from the algorithm, a dispersion analysis of localized water globules was carried out, supplemented with a frequency diagram describing the ratio of the size and number of particles found. The evaluation of the quality of the results of the work of the intelligent algorithm in comparison with the manual method on the verification microphotographs and the comparison of two empirical distributions allow us to conclude that the model based on the CNN can be verified and accepted for use in the search for particles in a fluid medium. The accuracy of the model was AP@50 = 89% and AP@75 = 78%.

Prediction of Mechanical Properties of Highly Functional Lightweight Fiber-Reinforced Concrete Based on Deep Neural Network and Ensemble Regression Trees Methods.

Currently, one of the topical areas of application of artificial intelligence methods in industrial production is neural networks, which allow for predicting the performance properties of products and structures that depend on the characteristics of the initial components and process parameters. The purpose of the study was to develop and train a neural network and an ensemble model to predict the mechanical properties of lightweight fiber-reinforced concrete using the accumulated empirical database and data from construction industry enterprises, and to improve production processes in the construction industry. The study applied deep learning and an ensemble of regression trees. The empirical base is the result of testing a series of experimental compositions of fiber-reinforced concrete. The predicted properties are cubic compressive strength, prismatic compressive strength, flexural tensile strength, and axial tensile strength. The quantitative picture of the accuracy of the applied methods for strength characteristics varies for the deep neural network method from 0.15 to 0.73 (MAE), from 0.17 to 0.89 (RMSE), and from 0.98% to 6.62% (MAPE), and for the ensemble of regression trees, from 0.11 to 0.62 (MAE), from 0.15 to 0.80 (RMSE), and from 1.30% to 3.4% (MAPE). Both methods have shown high efficiency in relation to such a hard-to-predict material as concrete, which is so heterogeneous in structure and depends on many factors. The value of the developed models lies in the possibility of obtaining additional useful information in the process of preparing highly functional lightweight fiber-reinforced concrete without additional experiments.